I have a question regarding the use of rma.mv (metafor) in a multivariate meta-analysis, in particular how to handle random factors. I have outcomes that vary not only between trials but also between subjects. In other words, there are outcomes that have been measured twice in one subject using different methods (mRNA, protein). This is true for only a few studies, while others only report one method.
So far, I have 'trial' as a random variable. If I add 'subject' as another random variable, I receive an error message in larger datasets saying that the optimizer could not converge. What would you suggest?
Please find attached the code and a test dataset. Thank you very much in advance. Lisa
yi <- dat1$SMD
V <- dat1$Variance
W<-dat1$NoPatients
res <- rma.mv(yi, V, W,mods = ~ Gen - 1, random =~ Gen | trial,struct="UN", data=dat1)
res
Multivariate Meta-Analysis Model (k = 53; method: REML)
Variance Components:
outer factor: trial (nlvls = 11)
inner factor: Gen (nlvls = 4)
estim sqrt k.lvl fixed level
tau^2.1 0.2331 0.4828 5 no NDUFA1
tau^2.2 0.0009 0.0303 10 no NDUFS1
tau^2.3 0.0147 0.1212 17 no NDUFV1
tau^2.4 0.0595 0.2440 21 no NDUFV2
rho.NDUFA rho.NDUFS rho.NDUFV1 rho.NDUFV2 NDUFA NDUFS NDUFV1 NDUFV2
NDUFA1 1 0.5679 0.9493 0.9989 - no no no
NDUFS1 0.5679 1 0.2804 0.6060 1 - no no
NDUFV1 0.9493 0.2804 1 0.9335 1 4 - no
NDUFV2 0.9989 0.6060 0.9335 1 2 4 5 -
Test for Residual Heterogeneity:
QE(df = 49) = 119.9290, p-val < .0001
Test of Moderators (coefficient(s) 1:4):
QM(df = 4) = 11.8456, p-val = 0.0185
Model Results:
estimate se zval pval ci.lb ci.ub
GenNDUFA1 0.4482 0.3377 1.3272 0.1845 -0.2137 1.1100
GenNDUFS1 -0.1999 0.0641 -3.1211 0.0018 -0.3255 -0.0744 **
GenNDUFV1 0.0145 0.0779 0.1857 0.8527 -0.1382 0.1671
GenNDUFV2 -0.0109 0.1272 -0.0854 0.9319 -0.2601 0.2383
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Aucun commentaire:
Enregistrer un commentaire